Over the past two years, I’ve downloaded all papers related to GEO, AEO, and AI search — 41 in total.

I read through them one by one and gained many new insights and inspirations.

All 41 papers have been pushed to a GitHub repository. Feel free to download them — the link is at the end.

Here are some key insights:

1. GEO is not a replacement for SEO

   • SEO addresses whether content can be retrieved, indexed, and included in the candidate set.
   • GEO addresses whether, after entering the candidate set, content can be cited, absorbed, and presented in AI answers. The two are additive.

2. Traditional search cares about ranking; AI search cares about answer participation

   • The old question: Where do I rank?
   • The new question: Am I cited? Where is the citation? Does my content actually influence the answer? Will users still click through?

3. “Being cited” ≠ “Having influence”

   • Some sources are merely listed in citations, with the AI answer barely using them.
   • What truly matters is citation absorption: whether your content contributes definitions, data, comparisons, steps, arguments, or answer structure.

4. The core of GEO is not turning articles into FAQs

   • Multiple papers show that Q&A format alone is not consistently effective.
   • What really works is an evidence container: clear conclusions, clear structure, high factual density, semantic alignment, verifiability, and extractability.

5. AI prefers structured content

   • Heading hierarchy, paragraphs, lists, tables, semantic HTML, structured data, update timestamps, metadata, etc., all affect whether AI understands, cites, and absorbs your content.
   • Content structure itself is a GEO signal.

6. A single AI search test has no decision value

   • The same question can yield different results across platforms, languages, prompt styles, time, and model versions.
   • GEO monitoring must be repeated, recording citation count, citation depth, answer share, position, and stability.

7. Different AI search engines are not the same

   • ChatGPT, Google AI Overviews, Perplexity, and Gemini have very different citation logics.
   • Some cite broadly, some cite less but absorb deeply. So GEO testing cannot rely on a single platform.

8. Third-party authority matters more than brand self-promotion

   • AI search clearly favors earned media, authoritative media, reviews, encyclopedias, industry reports, and trusted third-party sources.
   • Brand websites are important, but not enough.
   • Brands need to build both on-site evidence and off-site authority.

9. The boundary between GEO and black-hat manipulation is clear

   • Optimizing structure, evidence, readability, and factual density is white-hat.
   • Hidden instructions, prompt injection, fabricating facts, defaming competitors, and manipulating model ranking are black-hat.
   • AI search optimization will inevitably bring new security governance challenges.
   • But for brands, sticking to white-hat GEO from the start is the sustainable, healthy path.

10. Ultimately, GEO is not a tactic but a systematic capability

    • It includes: intent research, evidence assets, structured content, third-party authority, multi-platform monitoring, stability assessment, and security boundaries.
    • Those who continuously build this system will remain visible in AI search over the long term.

GitHub repository for the 41 papers: https://github.com/yaojingang/geo-citation-lab/tree/main/02-geo-aeo-ai-search-papers…

yaojingang/geo-citation-lab

Source: https://github.com/yaojingang/geo-citation-lab

GEO Citation Lab

GEO Citation Lab is a public repository for GEO research, positioned as:

• GEO Experiment Data Reports: Research on search triggering, citation sources, and page absorption based on ChatGPT, Google AI Overview / Gemini, and Perplexity.
• GEO / AEO / AI Search Paper Collection: Ongoing collection of papers related to generative search, AEO, GEO, AI search citation mechanisms, and manipulation risks.

The repository stores not just opinion articles, but verifiable data, scripts, reports, and paper materials in one place, facilitating secondary analysis, citation, and future expansion.

Start Here

Live Site: https://yaojingang.github.io/geo-citation-lab/

Repository Structure

Experiment Data Report Snapshot

The experiment primarily answers three questions:

• What kinds of questions most easily trigger AI to search the web?
• Which source websites does AI search prefer most?
• What types of pages are deeply absorbed by AI, rather than just “name-dropped” as citations?

Casual users can start with QUICK_REPORT.md. For the full reasoning, read final_report.md or final_report.pdf.

Paper Collection Snapshot

The new paper collection comes from the GEO_AI搜索_AEO_论文合集 (GEO/AI Search/AEO Paper Collection). Original batch directories have been merged into 7 topic directories, totaling 41 PDFs:

The full paper list is available at 02-geo-aeo-ai-search-papers/README.md. Two copies of GEO_AI搜索_AEO_论文整理说明.docx from the source directory had identical content; the repository retains one copy after SHA-256 deduplication, and the original 论文清单.csv is preserved.

How to Read

1. Start with 01-geo-experiment-data-report/QUICK_REPORT.md to quickly grasp the experiment conclusions.
2. Then read 01-geo-experiment-data-report/04-repet/final_report.md for the full methodology, charts, and chapter arguments.
3. Open 01-geo-experiment-data-report/02-data/features_all_platforms_72.csv to filter the fields you care about.
4. Read 02-geo-aeo-ai-search-papers/README.md and navigate into topic directories to access paper PDFs.

Running the Public Repository

This repository has been modified to read API keys from environment variables, avoiding private keys stored directly in GitHub.

cd 01-geo-experiment-data-report
cp .env.example .env

Common re-run commands:

cd 01-geo-experiment-data-report/03-pipeline
python3 analyze_influence.py \
  --input ../02-data/features_all_platforms_72.csv \
  --output ../04-repet/citation_influence_report.md

cd 01-geo-experiment-data-report/04-repet
python3 build_self_contained_html.py